Processors
Intel® Processors, Tools, and Utilities
14405 Discussions

Xeon X3470 and PCI Express lanes limitation

idata
Employee
3,330 Views

I recently purchased a used Xeon X3470 to replace my Core i5-760. I wanted to upgrade to the i7 and since the X3470 is pretty well identical to the i7-870 but cheaper plus can run 8 threads rather than 4 I thought it was a better buy. All the forums I read did not mention any major difference except for the ECC support on the Xeon and my motherboard listed it as a supported CPU. So I decided to buy it. The X3480/i7-880 is only ~3% faster so not worth the premium in price for my budget.

However when doing a side by side product comparison I spotted a difference that concerns me.

The Xeon CPUs have a maximum # of PCI Express Lanes of 1 while the i7 equivalents have a maximum # of PCI Express Lanes of 16.

Does this mean that the graphics performance with a PCIe graphics card and Xeon CPUs will be inferior to the equivalent i7?

What exactly does this difference of maximum # of PCI Express Lanes imply?

Thanks for taking the time to respond.

0 Kudos
5 Replies
IUman
Honored Contributor II
1,109 Views

Hello JamesKIWI,

Thank you for joining the Intel communities.

The Intel® core i7-880 and the Intel® Xeon® Processor X3470 support x16 graphics cards. The difference here is that the Xeon processor will send the information using one lane and the i7 processor will use 16 lanes so it will have better bandwidth.

Best regards,

Ivan

0 Kudos
idata
Employee
1,108 Views

Precisely. From what i read, a PCIE x16 Laned slot has a maximum bandwidth of 16GB/S under specification version 3.0. The more lanes that the hardware can utilize, the more bandwidth that will be available. Possible configurations can be X4, X8 and a max of X16 processing lanes per slot.

CPU's as well as Motherboard chip sets can have processing limitations of doing read/writes through out the PCIE lanes. You can have a motherboard that supports SLI/Crossfire for video processing in two PCIE lanes at Full X16 mode. But depending on the CPU you have, It can either do 2 in X8 lane for each bus or 2 in X16 lane for each bus.

From what i have been reading CPU's that do 32 or above lanes in processing can do X16 lanes in both slots. More mainstream variants like the Core i5 would be throttled down to two slots in X8 lane mode for processing. Each lane can process about 1GB/s so X4 would draw 4 GB/s X8 = 8GB/s and so forth. PCIE does simultaneously read/write. However i am not sure if the bandwidth is split evenly between each operation cycle, or if it does 1gig read followed by a 1 gig write. I do hope this helps.

0 Kudos
IUman
Honored Contributor II
1,108 Views

Thanks GreeneMan for that information you kindly provided, the i7 processors are used like when you want to use a video card for gaming and they are faster and the Intel® Xeon processor are used more for heavy applications like server systems that run multiple and heavy applications.

0 Kudos
idata
Employee
1,108 Views

Thanks to you both for that useful information.

In order to check my understanding, if I am running an older generation NVIDIA GTX 460 which supports PCIe 2.0, For graphic intensive work like video editing and gaming, I would be better off sticking with the i5-760 with 16 PCIe lanes and no hyper threading rather than upgrade to the Xeon 3470 with one PCIe lane and double the amount of threads and a better performance (27% better relative performance according to CPUupgrade.com). .

Because the i5-760 can push 500Mb/s of video data via 16 lanes (16 X 500Mb/se theoretical maximum) whereas the Xeon can only handle 500Mb/sec via one lane.

Of course going from the i5-760 to the i7-870 or i7-880 would be a good choice if I could afford it.

0 Kudos
idata
Employee
1,108 Views

For graphics processing yes, the i5 760 may be better. The Xeon could have more data processing power in terms of compute and memory bandwidth from a north bridge pov. I am assuming because i cannot find much on the chip other than whats listed on the intel site.

Applications can have processing bottle necks too which is something to be aware of. I cannot find the compute "Transistor density" on some of these chips. The xeon x3470 being one of them unfortunately. The i5 from what i can find should in theory have 774 million. This is important to know because my theory is that aside from the compute instruction set, SIMD engines like SSE 2,3 and so forth. If you clock two different CPU's at the same clock rate say 3GHZ, one will be faster than the other even on the same clock. For example a Lynnfield i5 over a Haswell Core i7. More electrical circuits in bit processing and calculation.

For Lynnfield family, the only difference i see apart from the i5 and i7 is cache memory capacity, turbo which is just auto clocking, and HT. The i7 having 1 extra logical thread over each of the 4 cores. I would go with the i5 760 or i7 860 if you want/need the hyper threading. From what i can see the 870 has higher base clock rates and turbo over the 860.

But IMO best choice of cpu would be the i7 860 for price performance. Never know when you can take advantage of the HT or not. Some applications support it, others do not i am sure. Texture manipulation software might not use much cpu processing. Video editing will use considerably more. Video gaming on some engines are really heavy. Which would make me want to use a Xeon, but the south bridge PCIE lane limitation could be worse off than just using a higher end cpu that does less instruction thread calculations per clock.

0 Kudos
Reply